TamGen
Pocket-conditioned GPT chemical language model paired with docking/IC50 feedback loops to design tuberculosis ClpP protease inhibitors.
The loop
- Proposes
A 3D-pocket-encoder + chemical-GPT decoder proposes SMILES structures conditioned on the target protein binding pocket and seed scaffolds.
- Judged by
RDKit synthetic accessibility, physics-based molecular docking, and wet-lab enzymatic IC50 and whole-cell M. tuberculosis assays evaluate candidates.
- Steers
Top-scoring docked and assayed hits seed iterative scaffold refinement rounds.
Headline result
Designed novel Mycobacterium tuberculosis ClpP inhibitors validated in wet-lab assays with strong enzymatic inhibition (IC50 = 0.9 uM).
What this does not prove
Requires a known 3D crystal structure of the target binding site to encode pocket geometry.
How it generalizes the harness
Physics docking filters and wet-lab enzymatic IC50 assays judge and steer pocket-conditioned molecular generation.
Classification
| What it proposes | Molecule |
|---|---|
| Generative model | 3D geometric pocket encoder + chemical GPT decoder |
| Algorithmic authority | Molecular docking engine and wet-lab IC50 enzymatic assay |
| Feedback | Score |
| What gets updated | Population of candidates and prompt context |
| When it runs | At inference time |
| Validity | Approximate physics (simulation, not lab results) |
| Evidence | Experiments reported |
| Lineage | Original system |